A Hybrid Neural Network Based on Bayesian Optimization for Non-Intrusive Load Disaggregation

Non-intrusive load disaggregation is an important technology for the smart grid to realize demand-side refinement management. Non-intrusive load disaggregation refers to the decomposition of the operating power of a single appliance from the total power. Considering the strong correlation of household loads in time series, a hybrid neural network load disaggregation method based on Bayesian optimization is proposed in this paper. The hybrid neural network is used to extract features in power sequences and Bayesian optimization is used to find a set of hyperparameter combinations suitable for hybrid neural networks. The method extracts deeper information, obtains a more flexible perceptual field, and improves the performance of the model. Finally, the constructed algorithm has experimented on the public dataset UK-DALE. The experimental results show that the MAE and F1 evaluation metrics are improved by 18.4% and 8.1%, respectively, which proves the superiority of the method in this paper.